Lesson 3

Decision Trees
 Expected Learning Outcomes

1. Develop accurate and useful decision trees

2. Revise probabilities using Bayesian

analysis
 Decision Trees
• Any problem that can be presented in a decision
table can also be graphically represented in a
decision tree
• Decision trees are most beneficial when a
sequence of decisions must be made
• All decision trees contain decision points or nodes
and state-of-nature points or nodes
– A decision node from which one of several alternatives
may be chosen
– A state-of-nature node out of which one state of nature
will occur
 Five Steps to
Decision Tree Analysis

1. Define the problem

2. Structure or draw the decision tree
3. Assign probabilities to the states of nature
4. Estimate payoffs for each possible
combination of alternatives and states of
nature
5. Solve the problem by computing expected
monetary values (EMVs) for each state of
nature node
 Structure of Decision Trees

• Trees start from left to right

• Represent decisions and outcomes in
sequential order
• Squares represent decision nodes
• Circles represent states of nature nodes
• Lines or branches connect the decisions
nodes and the states of nature nodes
 Example:
Thompson Lumber Company

Step 1 – Define the problem

■ Expand by manufacturing and
marketing a new product, backyard
storage sheds
Step 2 – List alternatives
■ Construct a large new plant
■ A small plant
■ No plant at all
Step 3 – Identify possible outcomes
■ The market could be favorable or
unfavorable
Thompson Lumber Company

Step 4 – List the payoffs

■ Identify conditional values for the
profits for large, small, and no plants
for the two possible market conditions
Step 5 – Select the decision model
■ Depends on the environment and
amount of risk and uncertainty
Step 6 – Apply the model to the data
■ Solution and analysis used to help the
decision making
 Thompson Lumber Company Payoff

STATE OF NATURE

FAVORABLE UNFAVORABLE
ALTERNATIVE MARKET (Php) MARKET (Php)
Construct a large plant 200,000 –180,000

Construct a small plant 100,000 –20,000

Do nothing 0 0

Table 3.1
 Thompson’s Decision Tree
A State-of-Nature Node
Favorable Market
A Decision Node
1
Unfavorable Market
u ct nt
r
n st Pla
e
Co arg
L Favorable Market
Construct
Small Plant 2
Unfavorable Market
Do
No
th
in
g
Figure 3.2
 Thompson’s Decision Tree payoffs

EMV for Node 1 = = (0.5)(P200,000) + (0.5)(–P180,000)

P10,000
Payoffs
Favorable Market (0.5)
P200,000
Alternative with best
EMV is selected 1
Unfavorable Market (0.5)
ct nt –P180,000
r u
n st Pla
e
Co arg
L Favorable Market (0.5)
P100,000
Construct
Small Plant 2
Unfavorable Market (0.5)
–P20,000
Do
No
th EMV for Node 2 = = (0.5)(P100,000)
in
g P40,000 + (0.5)(–P20,000)
Figure 3.3
P0
 Thompson’s Complex Decision Tree
(with Market survey)
First Decision Second Decision Payoffs
Point Point
Favorable Market (0.78)
P190,000
t 2 Unfavorable Market (0.22)
lan
P –P190,000
ge
Lar Small Favorable Market (0.78)
P90,000
5) 3 Unfavorable Market (0.22)
0 .4 Plant –P30,000
(
e y ts e
u rv sul abl No Plant
–P10,000
S Re or
v
1 Surv Fa Favorable Market (0.27)
P190,000
y

e
ve

Re y ( t 4 Unfavorable Market (0.73)

Ne su 0.5 lan
ur

P –P190,000
ge
tS

5)
ga lts
tiv Lar Small Favorable Market (0.27)
P90,000
ke

e 5 Unfavorable Market (0.73)

ar

Plant –P30,000
tM
uc

No Plant
–P10,000
nd
Co

Do Favorable Market (0.50)

Not P200,000
Con nt 6 Unfavorable Market (0.50)
d uct Pla –P180,000
ge
Sur
vey Lar Small Favorable Market (0.50)
P100,000
Plant 7 Unfavorable Market (0.50)
–P20,000
No Plant
P0
Figure 3.4
 Thompson’s Complex Decision Tree

1. Given favorable survey results,

EMV(node 2) = EMV(large plant | positive
survey)
= (0.78)(P190,000) + (0.22)(–P190,000) =
P106,400
EMV(node 3) = EMV(small plant | positive
survey)
= (0.78)(P90,000) + (0.22)(–P30,000) = P63,600
EMV for no plant = –P10,000
2. Given negative survey results,
EMV(node 4) = EMV(large plant | negative
survey)
= (0.27)(P190,000) + (0.73)(–P190,000) =
–P87,400
EMV(node 5) = EMV(small plant | negative
survey)
 Thompson’s Complex Decision Tree

3. Compute the expected value of the

market survey,
EMV(node 1) = EMV(conduct survey)
= (0.45)($106,400) + (0.55)($2,400)
= P47,880 + P1,320 = P49,200

4. If the market survey is not conducted,

EMV(node 6) = EMV(large plant)
= (0.50)($200,000) + (0.50)(–P180,000) =
P10,000
EMV(node 7) = EMV(small plant)
= (0.50)(P100,000) + (0.50)(–P20,000) =
P40,000
EMV for no plant = $0
 Thompson’s Complex Decision Tree
First Decision Second Decision Payoffs
Point Point
P106,400 Favorable Market (0.78)
P190,000
nt Unfavorable Market (0.22)
Pla –P190,000

P106,40
g e
Lar SmallP63,600 Favorable Market (0.78) P90,000
5)

0
.4 Unfavorable Market (0.22)
0 Plant –P30,000
(
e y ts e
urv sul abl No Plant
–P10,000
S Re or
Su Fav –P87,400 Favorable Market (0.27)
rv P190,000
y

e
ve

Re y (
Ne su 0.5 ant
l Unfavorable Market (0.73)
ur

eP –P190,000
g
tS

5)
P2,400
ga lts r
La Small P2,400 Favorable Market (0.27)
tiv P90,000
ke

e Unfavorable Market (0.73)

ar

Plant –P30,000
M
ct

No Plant
du

–P10,000
on
P49,200
C

Do P10,000 Favorable Market (0.50)

Not P200,000
Con
d ant
l Unfavorable Market (0.50)
e P P40,000 –P180,000
P40,000

uct g
Sur
vey Lar Small
Favorable Market (0.50)
P100,000
Unfavorable Market (0.50)
Plant –P20,000
No Plant
P0
Figure 3.4
Expected Value of Sample Information
(EVSI)

• Thompson wants to know the actual value of

doing the survey
Expected value Expected value
with sample of best decision
EVSI = information, assuming without sample
– no cost to gather it information

= (EV with sample information + cost)

– (EV without sample information)

EVSI = (P49,200 + P10,000) – P40,000 = P19,200

 Sensitivity Analysis
■ How sensitive are the decisions to changes in the
probabilities?

■ How sensitive is our decision to the probability of a

favorable survey result?

■ That is, if the probability of a favorable result

(p = .45) where to change, would we make the same
decision?

■ How much could it change before we would make a

different decision?
 Sensitivity Analysis

p = probability of a favorable survey

result
(1 – p) = probability of a negative survey
result 1) = (P106,400)p +(P2,400)(1 –
EMV(node
p)
= P104,000p + P2,400
We are indifferent when the EMV of node 1 is the
same as the EMV of not conducting the survey,
P40,000

P104,000p + P2,400 =
P40,000
P104,000p = P37,600
p = (P37,600)/(P104,000) =
 Bayesian Analysis
■ Many ways of getting probability data
■ It can be based on
• Management’s experience and intuition
• Historical data
• Computed from other data using Bayes’
theorem
■ Bayes’ theorem incorporates initial
estimates and information about the
accuracy of the sources
■ Allows the revision of initial estimates
based on new information
 Calculating Revised Probabilities

■ In the Thompson Lumber case we used these four

conditional probabilities
P (favorable market(FM) | survey results positive) = 0.78
P (unfavorable market(UM) | survey results positive) = 0.22
P (favorable market(FM) | survey results negative) = 0.27
P (unfavorable market(UM) | survey results negative) = 0.73

■ The prior probabilities of these markets are

P (FM) = 0.50
P (UM) = 0.50
 Calculating Revised Probabilities

■ Through discussions with experts Thompson has

learned the following
■ He can use this information and Bayes’ theorem
to calculate posterior probabilities
STATE OF NATURE
RESULT OF FAVORABLE MARKET UNFAVORABLE MARKET
SURVEY (FM) (UM)

Positive (predicts P (survey positive | FM) P (survey positive | UM)

favorable market
for product) = 0.70 = 0.20

Negative (predicts P (survey negative | P (survey negative |

unfavorable FM) UM)
market for
product) = 0.30 = 0.80

Table 3.11
Calculating Revised Probabilities

■ Recall Bayes’ theorem is

where

For this example, A will represent a favorable

market and B will represent a positive survey
 Calculating Revised Probabilities

■ P (FM | survey positive)

■ P (UM | survey positive)

 Calculating Revised Probabilities

POSTERIOR PROBABILITY
CONDITIONAL
PROBABILITY P(STATE OF
P(SURVEY NATURE |
STATE OF POSITIVE | STATE PRIOR JOINT SURVEY
NATURE OF NATURE) PROBABILITY PROBABILITY POSITIVE)
FM 0.70 X 0.50 = 0.35 0.35/0.45 = 0.78
UM 0.20 X 0.50 = 0.10 0.10/0.45 = 0.22
P(survey results positive) = 0.45 1.00

Table 3.12
 Calculating Revised Probabilities

■ P (FM | survey negative)

■ P (UM | survey negative)

 Calculating Revised Probabilities

POSTERIOR PROBABILITY
CONDITIONAL
PROBABILITY P(STATE OF
P(SURVEY NATURE |
STATE OF NEGATIVE | STATE PRIOR JOINT SURVEY
NATURE OF NATURE) PROBABILITY PROBABILITY NEGATIVE)
FM 0.30 X 0.50 = 0.15 0.15/0.55 = 0.27
UM 0.80 X 0.50 = 0.40 0.40/0.55 = 0.73
P(survey results positive) = 0.55 1.00

Table 3.13
Performance Assessment/Task 1
(PAT 1)